Labor Unemployment Risk and Corporate Tax Avoidance

Transcription

1 Labor Unemployment Risk and Corporate Tax Avoidance Erik Devos Department of Economics and Finance College of Business Administration University of Texas at El Paso El Paso, TX Shofiqur Rahman Finance Department College of Business New Mexico State University Las Cruces, NM January 2015 Corresponding author: phone: (915) We thank Garry Braun, Bill Elliott, Jim Upson, Desmond Tsang and seminar participants at the University of Texas at El Paso for comments. The authors retain responsibility for any remaining errors.

2 Labor Unemployment Risk and Corporate Tax Avoidance ABSTRACT This paper examines whether firms are less income tax avoiding in order to mitigate workers exposure to unemployment risk. We use unemployment insurance (UI) benefit laws as a proxy for unemployment risk and multiple measures of tax avoidance. Given that tax avoidance is risky and costly for the firm and its employees, we argue that low state UI benefits increase labor unemployment risk and, hence influence firms tax strategies. Consistent with our main hypothesis, we find a negative relation between tax avoidance and unemployment risk. In additional analysis, we also find that this negative relation is more pronounced for firms in industries that are more labor intensive, firms with financial constraints, and industries in which labor has more bargaining power. Overall, our findings suggest that labor market frictions have implications for corporate tax policy. JEL classification: J31; J65; H25 Keywords: Unemployment risk; Compensating wage differentials; Unemployment insurance benefit; Tax avoidance.

3 Labor Unemployment Risk and Corporate Tax Avoidance 1. Introduction Tax expenditures of U.S. firms are not trivial. We find that the median effective tax rate of US firm during the period equals 31 percent. It is therefore not surprising that a substantial amount of literature exists that suggests that firms may aggressively attempt to push their taxes down (e.g., Chyz, Leung, Li, and Rui 2013; Richardson, Taylor, and Lanis 2013; Lanis and Richardson 2011; Lanis and Richardson 2012; Chen, Chen, Cheng, and Shevlin 2010; Armstrong, Blouin, and Larcker 2012). 1 However, this tax avoidance may not be without costs. Other than potential costly IRS actions (e.g., Rego and Wilson 2009; Weisbach 2002), there may be stakeholders that could be affected negatively (e.g., Desai and Dharmapala 2008; Desai and Dharmapala 2009; Chen et al. 2010). In our paper, we focus on the relation between tax avoidance of the firm and one specific group of stakeholders that may be negatively affected by relatively high levels of tax avoidance: employees. To be precise, we study the relation between tax avoidance and labor unemployment risk. Empirically, we test whether firms located in states with higher unemployment risks (i.e., low unemployment insurance (UI) benefits) engage in less tax avoiding activities than do firms located in states with lower labor unemployment risk. 2 Prior literature suggests that workers perceptions of unemployment risk impact firm policy on layoffs and wage setting and reduce labor supply (e.g., Hamermesh and Wolfe 1990; Li 1986; Topel 1984; Topel 1983). In addition, Agrawal and Matsa (2013) argue that labor 1 See also Hanlon and Heitzman (2010) for a review of the tax literature. 2 To our knowledge, there exists only one paper that investigates a somewhat similar issue. Chyz et al. (2013) look at unionization rates and tax aggressiveness. Specifically, that means that they investigate the effects that bargaining power of employees may have on tax aggressiveness. We look at one of the underlying reasons that may cause employees and firms to reduce tax aggressiveness: unemployment risk. Moreover, we analyze a sample of over 60,000 firm-years, whereas their paper covers a much smaller sample. 3

4 market frictions (i.e., unemployment risk) are important factors in the corporate decision making process. 3 Combined, this would suggest that less tax avoiding behavior may, at least partially, be a reflection of a firms effort to mitigate workers exposure to unemployment risk. Prior literature also suggests that workers require higher compensating wage differentials (i.e., a premium in wages and benefits) when they perceive unemployment risk (e.g., Topel 1984; Abowd and Ashenfelter 1981). If this is the case, firms may choose conservative corporate policies, such as less tax avoidance, in order to minimize unemployment risk, and indirectly reduce wages and benefits. So, we hypothesize that, in the presence of labor unemployment risk, firms will exhibit less tax avoiding behavior. However, measuring unemployment risk is a challenging task. We follow Agrawal and Matsa (2013) and use state unemployment insurance (UI) benefit laws and examine the relation with tax avoidance of US firms over the period 1987 to Topel (1984) argues that higher UI benefits make layoffs less costly and reduces demand of workers for compensating wage differentials. Hence, firms should avoid less tax when unemployment risk is relatively high (i.e., the state has lower UI benefits). Based on state UI benefits, we calculate a proxy for unemployment risk. Specifically, we use the log of the maximum weekly benefit amounts multiplied by the maximum duration in weeks associated with these benefits. Higher Log max total benefit indicates lower unemployment risk. To test whether a relation between unemployment risk and tax avoidance exists, we employ four measures of tax avoidance: (1) the effective tax rate (ETR), measured as the total tax expense divided by pre-tax book income and; (2) the cash effective tax rate (CETR), measured as the cash tax paid divided by pre-tax book 3 They find that companies choose conservative financing policies in order to mitigate unemployment risk. This finding is consistent with theoretical work by Titman (1984) and Berk, Stanton, and Zechner (2010). A somewhat related stream of literature suggests that labor unions have incentives and risk preferences that manifest itself in conservative financial reporting (e.g., Faleye, Mehrotra, and Morck 2006; Leung, Li, and Rui 2012; Chyz et al. 2013). 4

5 income; (3) the discretionary book tax differences (DTAX) of Frank, Lynch, and Rego (2009), measured as the residual from a regression of permanent book-tax differences on nondiscretionary sources of those differences; and (4) the tax shelter prediction score (SHELTER) of Wilson (2009) 4. Our main empirical evidence suggests that firms located in states with higher unemployment risk (i.e., lower UI benefits) exhibit more less tax avoiding behavior than firms located in lower labor unemployment risk states. These results are consistent across all four measures of tax avoidance, are highly statistically significant, and have economic significance. We find that there is substantial economic impact of UnempRisk on ETR, CETR, DTAX, and SHELTER. Using the mean level of pre-tax income of $113 million for our sample and a change in Log max total benefit from the third to the first quartile, the impact of an increase in UnempRisk translates into an increase between $4.69 million and $6.00 million of tax expenditures for ETR and CETR, respectively. In order to address unobservable time-invariant firm characteristics, we run firm fixed effect model. Fixed effects estimation maintains separate intercepts for each firm and capture changes in unemployment risk and tax avoidance across time rather than cross-sectional correlations. Our results show that the relation between unemployment risk and tax avoidance holds within firms and further validating our main hypothesis. We also analyze the impact of labor intensity, financial constraints, and bargaining power. As expected, our results support our additional empirical predictions that the relation between tax avoidance and labor unemployment risk is stronger in industries that are more labor intense, firms with financial constraints, and industries with higher labor bargaining power. 4 All variables are defined in Appendix A. 5

6 Finally, our results are robust to the use of an alternative proxy for unemployment risk, exclusion of firms with foreign sales, exclusion of loss firms, exclusion of big firms, and exclusion of industries that have a dispersed labor force. Our paper makes several contributions to the existing literature on labor friction and corporate policy. First, our paper attempts to answer calls by Scholes, Wolfson, Erickson, Maydew, and Shevlin (2005) and Desai and Dharmapala (2006) for more research on tax avoidance. Their calls are specific, in that they ask for more research of tax research in an agency context. Although employees cannot perfectly contract with management, they do bestow decision making power regarding tax policy on managers, while at the same time demanding higher compensating wage differentials, when facing unemployment risk. Specifically, workers can indirectly make managers consider unemployment related costs in the tradeoff analysis of tax aggressiveness. Our paper adds an additional non-tax cost (labor unemployment risk) to this agency dynamic. Second, Agrawal and Matsa (2013), in their study of labor unemployment risk and debt policy, argue that labor market friction is a vital factor in the corporate environment and call for more research on similar corporate policies. We attempt to fill this void in the literature by providing empirical evidence suggesting that firms may avoid less tax in order to mitigate workers exposure to unemployment risk, thereby reducing the demand for higher wage differentials. Finally, by investigating a new non-tax determinant of tax avoidance (i.e., a macro variable that measures labor unemployment risk) we add to the existing literature on crosssectional determinants of tax avoidance (e.g., Gupta and Newberry 1997; Dyreng, Hanlon, and Maydew 2008, Graham and Tucker 2006; Shevlin 2001). The remainder of this paper is structured as follows: In section 2, we discuss how our paper relates to the existing literature and develop our main hypothesis. Section 3 describes the 6

7 data, our measures of tax avoidance and unemployment risk, and our research design. In section 4, we discuss our main empirical findings. Section 5 discuss the results of additional analysis. In section 6, we discuss our robustness checks. Finally, section 7 summarizes our findings and concludes the paper. 2. Related literature and hypothesis development Existing literature on tax avoidance, motivated by the Scholes et al. (2005) framework focuses on outcomes related to tax avoidance in which rational managers will engage in tax avoidance activities as long as the marginal benefits outweigh the marginal costs. One of the obvious benefits of tax avoidance is the tax savings that may potentially increase the amount of cash flows available to the firm. Managers may benefit from tax avoidance activities as well, either directly or indirectly through compensation. In addition, they may use these activities for rent extraction using masked and opaque tax avoidance (Desai and Dharmapala 2006). Some examples of empirical work that centers on costs of tax avoidance or tax planning implementation costs are Mills and Newberry (2001), Guenther, Maydew, and Nutter (1997), and Rego and Wilson (2012). Existing literature also suggests that firm s tax avoidance is risky (e.g., Rego and Wilson 2012; Guenther, Matsunaga, and Williams 2013; Brown, Drake, and Martin 2012). In light of the fact that these benefits and costs may accrue to not only the shareholders of the firm it is not surprising that a growing body of research investigates tax outcomes for different stakeholders. Examples of recent studies in this area examine tax avoidance from the perspective of debt holders (Kim, Li, and Li 2010; Lisowsky, Mescall, Novack, and Pittman 7

8 2011) and taxing authorities (Desai, Dyck, and Zingales 2007). Finally, there is also evidence that labor unions may affect tax avoidance (Chyz et al. 2013). They find that labor unionization rates are negatively associated with tax avoidance at both the firm and the industry level. These last findings are part of a broader stream of literature that provides empirical evidence of the impact that labor market frictions can have on corporate decision making. Most related to our work, Agrawal and Matsa (2013) argue that changes in UI benefits can provide meaningful shocks to the costs to workers. Similar studies that also examine the interaction of labor economics and finance include Benmelech, Bergman, and Enriquez (2012), Chen, Kacperczyk, and Ortiz-Molina (2011), and Matsa (2010) How labor unemployment risk impact tax avoidance Involuntary unemployment imposes costs on the labor force. Examples of such costs are job search costs (Mortensen and Pissarides 1994; Mortensen 1986; Diamond 1982), layoff discouragement effects (Jahoda 1982), a limited supply of matched job opportunities (Lazear 2003), and other possible labor market frictions. These unemployment related costs lead workers to require compensation in the form of higher wages, additional benefits, and quality working conditions. These direct or indirect payments are usually labeled compensating wage differentials. Given that firms cannot guarantee workers with insurance covering these costs, a number of studies argue that firms must compensate workers ex ante to bear such unemployment risk (e.g., Hamermesh and Wolfe 1990; Li 1986; Topel 1984; Abowd and Ashenfelter 1981). Firms compensation benefits payments, therefore, should increases with the probability of unemployment, the degree of workers risk averseness, and the duration of joblessness. 8

9 Previous studies provide empirical evidence on the importance of compensating wage differentials for unemployment risk. Abowd and Ashenfelter (1981) show that compensating wage differentials vary across industries and account for about 14 percent of total wages. Hamermesh and Wolfe (1990) argue that percent of the industry wage differentials can be explained by the change in unemployment risk. Compensating wage differentials seem to impact firms corporate policies. Besides, consistent with Hanlon and Heitzman (2010), Rego and Wilson (2012), and Badertscher, Katz, and Rego (2013), we also argue that tax avoidance is a risky activity for firms and managers who, if engaged in such activity, bear significant costs that include, but are not limited to, fees paid to tax experts for tax planning, time devoted to the resolution of tax audits, reputational penalties, and penalties paid to tax authorities (IRS) if detected. As tax avoidance makes the firm risky in the eyes of the labor force, whether to engage in this activity (and by how much) depends on the costs and benefit tradeoff. Hence, unemployment risk raises the costs that are required to compensate workers for bearing greater exposure to this risk. The following equation presents a simplified view of avoiding tax: Net benefits (tax avoidance) = Gross benefits Costs (direct and indirect) Wage differentials (unemployment risk) (1) Based on Equation (1), firms engage in tax avoidance activities as long as the benefits exceed the costs. In this paper, we argue that when workers are already exposed to unemployment risk, further risky corporate activity (e.g., tax avoidance) is likely to increase wage differentials, reducing the net benefits of tax avoidance. 5 Given that the level of risk 5 The demand for higher wage differentials when firms engage in risky corporate activities (e.g., tax avoidance) does not require that workers observe firms tax avoidance decisions. Workers may recognize these risks from 9

10 aversion is an important input into the model of optimal tax evasion (Crocker and Slemrod 2005), we expect labor unemployment risk to increase the marginal costs of tax avoidance, leading managers to avoid less tax. A similar argument is that workers will more likely demand higher wages and costly benefits if they perceive cash flows as riskier due to managers investment in tax avoidance. Consistent with Agrawal and Matsa (2013) (managers use less debt to assure laborers of lower bankruptcy risk), Chen et al. (2010) (managers engage in less tax aggressiveness because outside investors suspect increased agency costs), and Chyz et al. (2013) (managers engage in less tax aggressiveness because of the demand from labor unions of higher compensation), managers are more likely to reduce tax avoidance when labor unemployment risk is higher so as to reduce the need to compensate for higher wage differentials Hypothesis The discussion above suggests that when laborers perceive high unemployment risk, they require a higher level of wage differentials, increasing the costs of tax avoidance. This argument leads us to the following hypothesis. MAIN HYPOTHESIS: Firms located in states with higher unemployment risk (i.e., lower UI benefits) exhibit less tax avoiding behavior. A finding of less tax avoidance by firms located in states with high unemployment risk is consistent with the idea that unemployment risk can constrain firm behavior and decrease the net informative signals from coworkers, management, the media, the regulatory authorities, and economic environment. To illustrate, Brown and Matsa (2012) find that job seekers are able to perceive the financial situation of a firm. 10

11 benefits of tax avoidance. We test this hypothesis by conducting cross-sectional analyses at the firm-year level. 3. Data and research design 3.1. Measures of tax avoidance The dependent variable in our empirical tests is corporate tax avoidance. To be consistent with prior literature (i.e., Chen et al. 2010; Rego and Wilson 2012; Chyz et al. 2013), we use multiple measures of tax avoidance. All measures are calculated using data from the Compustat annual files. The first measure of tax avoidance is the effective tax rate (ETR), which is computed as total tax expense (Compustat data#16) divided by pretax income (data# 170). The second measure is the cash effective tax rate (CETR), which is computed as cash taxes paid (data# 317) divided by pretax income (data# 170). Higher values of ETR and CETR represent lower tax avoidance. We winsorize both these measures at [0,1]. While ETR is a simple measure, Dyreng et al. (2008) argue that CETR is a more accurate variable in measuring a firm s ability to maintain persistent tax avoidance strategies. ETR and CETR can be considered measuring benign forms of tax avoidance. Guenther, Matsunaga, and Williams (2013) argue that firms with persistently lower cash ETR are firms that avoid tax through safe tax positions. We therefore include two additional proxies (e.g., DTAX and SHELTER) that are considered measures of aggressive tax avoidance. Our third measure is discretionary book tax differences (DTAX) of Frank, Lynch, and Rego (2009). DTAX is the residual from a regression of permanent book-tax differences on non-discretionary sources of those differences. Our fourth measure of tax avoidance proxy is a tax shelter prediction score 11

12 (SHELTER, as calculated in Wilson, 2009) 6. Higher values for DTAX and SHELTER represent more tax avoidance Proxy for labor unemployment risk Given that a precise measure or appropriate proxy for labor unemployment risk is scarce in the existing literature, measuring workers exposure to unemployment risk is a challenging task. Following Agrawal and Matsa (2013), we use the log of the maximum total benefit, unemployment insurance (UI), as a proxy for unemployment risk. We calculate the maximum total benefit (Log max total benefit) as: Log of Maximum Total Benefit = log (Maximum weekly benefit amount Maximum number of weeks) (2) We collect information on UI benefits from the US Department of Labor s Significant Provisions of State UI Laws from 1987 through These annual publications contain detailed explanations of UI benefit schedules for each state. According to the publication, eligible claimants receive a weekly benefit payment for a specified number of weeks. The level and duration of UI benefit payment depends on workers employment history during a base period. From each publication, we collect the maximum weekly benefit amount and the maximum duration a claimant is eligible to enjoy the benefit. As Equation (2) shows, our main focus is on the product of the maximum weekly benefit amount and the maximum duration allowed. As an additional test, we use the maximum weekly benefit amount as a proxy for UI 6 Both DTAX and SHELTER can be used to isolate intentional tax avoidance, which is relatively risky. Details on how DTAX and SHELTER are computed are provided in Appendix A. 7 We start our sample period in 1987 because income tax paid (Compustat TXPD), which is used for calculating one of the measures of tax aggressiveness (CETR), is missing for years before

13 benefits. We limit our analysis to UI benefit amount in dollar (e.g., Log max total benefit and Log max weekly benefit). 8 State UI benefits can be considered a good proxy for unemployment risk. Topel (1984) argues that higher UI benefits make layoffs less costly and reduce workers demand for being compensated by their employers for facing unemployment risk. A number of studies show that UI compensation has economically meaningful effects on workers behavior and on aggregate labor supply (e.g., Topel and Welch 1980; Topel 1984; Meyer 1990; Meyer and Mok 2007; Gormley, Liu, and Zhou 2010). Agrawal and Matsa (2013) find that measures of UI benefits are reflected in the aggregate realized value of UI benefits paid by each state. They reported an elasticity of 0.9 between maximum total benefit and actual compensation payments. They also find that aggregate compensation payout is correlated with both the maximum weekly benefit amounts and the maximum duration in weeks Sample selection and descriptive statistics We test the association between corporate tax avoidance and labor unemployment risk at the firm-year level. We hand collect unemployment insurance benefit data from the US Department of Labors significant provisions of State UI Laws from 1986 to We also collect firms balance sheet and income statement data from the Compustat annual file for the period between 1986 and We match the Compustat sample with the UI benefit sample in two steps. First, our initial sample has 257,245 firm-year observations. We exclude financial firms (SIC ) and utilities (SIC ) because these firms are subject to 8 Maximum duration allowed has less variation across the states. The standard deviation is only as opposed to the and for Log max total benefit and Log max weekly benefit, respectively. 13

14 different regulations. As a result we delete 75,378 firm-years. For our regression models, we require non-missing values for the control variables (e.g., ROA, Leverage, Market-to-book, size, foreign income, PPE, Intangible, equity income in earnings, loss carry forward; 55,873 firmyears). We only keep sample firm-years if we are able to calculate our four tax avoidance measures (ETR, CETR, DTAX, and SHELTER). This requirement screens out 60,089 firm-years. Finally, we match state employment insurance data with Compustat data based on the state where a firm s headquarter is located. Company headquarters information for some firms is missing in Compustat. Because the unemployment risk measure for a firm is dependent on the state where it is located, we only keep those firm-years for which we are able to get the unemployment risk variable (Log max total benefit). This deletes 3,159 firm-years. We also delete 22 observation for which we are not able to calculate a state GDP growth rate and, as a result, there are 62,724 firm-years in our final sample. We describe the sample formation process in detail in Table 1, Panel A. In Panel B of the Table, we present how the sample is represented in different sample periods. It appears that our firm-years are not particularly clustered though time. In Panel C, we present all 2-digit SIC codes that comprise at least two percent of the firm-years in our sample. There are thirteen such industries which in aggregate contain about 33 percent of the sample. The 2-digit industry that is most represented is 73 - Business Services (13.8 percent), followed by 36 Electronic and Other Equipment (9.5 percent) The summary statistics of our tax avoidance proxies, unemployment risk, and control variables are presented in Table 2. The mean (median) of ETR, CETR, DTAX and SHELTER are 0.25 (0.31), 0.19 (0.10), 0.16 (0.05), and 1.29 (-0.50) respectively. The UnempRisk is the natural log of maximum weekly benefit amount times the maximum number of weeks. The average 14

15 UnempRisk is 9.05 (approximately $8,536 per week). 9 When it comes to the summary statistics of our control variables 10, we find that the mean (median) return on assets in our sample is (0.048). The mean (median) firm has approximately 20 percent (11 percent) debt in its capital structure. The mean (median) market value of equity, a proxy for size, is $115 ($118) million. The mean and median GDP growth rate in the sample is about 5 percent. To mitigate concerns that there may not be much variation in our benefit variables, we investigate how much variation there is in this variable across states and across time (within a given state). We do find a significant variation in unemployment benefits across states. In 2012, for example, the maximum total unemployment insurance benefit varies from $6,110 in Mississippi to $29,370 in Massachusetts. We replicate the figures of Agrawal and Matsa (2013) using our data in order to validate our datasets and graphically present the evolution of UI benefits over time across states. Figure (1) presents the percentage change in UI benefit by quinquennium. Each map depicts the quartile change in UI benefit for each state in that different time period (i.e., five years). The darker shades indicate a larger percentage change in benefits. States display a reasonable heterogeneity in UI benefit trends. In Figure (2), we plot the distribution of absolute changes in UI benefits. We find a 10 percent to 20 percent (absolute) change in UI benefits by US states in each five years. Although less prevalent, larger (absolute) benefit changes also occur. 9 The Log max weekly benefit is the natural log of the maximum weekly benefit allowance given to a qualified worker in an average state-year. The Log max duration is the natural log of maximum number of weeks a state provides unemployment benefits to a worker. In an unreported table, the average Log max weekly benefit is 5.78 (approximately $324 per week).the average Log max duration is 3.27 (approximately 26 weeks). 10 All the control variables are winsorized at 1 percent and 99 percent tails 15

16 3.5. Research design We use cross-sectional regressions to examine the relation between tax avoidance and unemployment risk. We estimate the following base model: TaxAvoidance it = α 0 + β 1 UnempRisk st 1 + β 2 ROA it + β 3 LEV it + β 4 NOL it + β 5 NOL it + β 6 FI it + β 7 PPE it + β 8 INTANG it + β 9 EQINC it + β 10 SIZE it 1 + β 11 MB it 1 + β 12 GDP_gr it + StateDummies + IndustryDummies + ε it (3) where TaxAvoidance it are tax avoidance proxies (i.e., ETR, CETR, DTAX and SHELTER); UnempRisk st 1 is our proxy for labor unemployment risk. This variable is the log of maximum weekly benefit amount times the maximum number of weeks a certain state provides unemployment benefit to a claimant 11. All these (and other) variables are measured at the firmyear level. We define our control variables as follows: ROA it is measured as operating income (data# 170 data# 192) scaled by lagged assets (data# 6); LEV it is measured as long-term debt (data# 9) scaled by lagged assets; NOL is an indicator variable coded as one if loss carry forward (data# 52) is positive as of the beginning of the year t; NOL is the change in loss carry forward (data# 52) scaled by lagged assets; FI it is foreign income (data# 273) for firm i in year t scaled by lagged assets; PPE it is the net property, plant, and equipment (data# 8) for firm i in year t scaled by lagged assets; INTANG it stands for intangible assets (data# 33) for firm i in year t scaled by lagged assets; EQINC it is measured as the equity income in earnings (data# 55) for firm i in year t scaled by lagged assets; SIZE it 1 is the natural log of market value of equity (data# 199 data# 25) for firm i at the beginning of year t; MB it 1 refers to the market-to-book 11 In all the regression models, we multiply log max total benefit variable by -1 in order to construct UnempRisk variable. This allows for easier interpretation of our results. 16

17 ratio measured as market value of equity (data# 199 data# 25) divided by book value of equity (data# 60) for firm i at the beginning of year t. GDP_gr it is the change in state GDP in year t and t-1. If firms are located in states with higher labor unemployment risk (i.e., lower UI benefits) and are less tax avoiding, we expect the coefficient on UnempRisk, β 1 to be positive for ETR and CETR and negative for DTAX and SHELTER. We also consider certain firm characteristics that may potentially play roles as determinants of corporate tax avoidance. To be consistent with prior studies (e.g., Mills 1998; Manzon and Plesko 2002; Rego 2003; Dyreng, et al. 2008; Frank, Lynch, and Rego 2009; Wilson 2009; Chen et al. 2010; Rego and Wilson 2012), we control for two sets of firm characteristics to ensure that our results are not driven either by fundamental differences among firms with different levels of unemployment risks that workers are facing or by differential book and tax treatment on capital investment. The first set of control variables is designed to capture firms profitability, leverage, and foreign operations are ROA, LEV, NOL, NOL, and FI. For example, more profitable firm and firms with lower levels of tax loss carry forwards tend to have higher effective tax rates. Similar logic can be applied to firms with foreign income. Graham and Tucker (2006) find that tax shelter firms have lower leverage ratios than their control firms, suggesting that leverage and tax shelter activity serve as substitutes. Therefore, leverage is expected to be negatively related to the effective tax rate. The second set of control variables is designed to capture differences in book and tax reporting, they are Plant, property, and equipment (PPE), Intangible assets (INTANG), and Equity income (EI). For example, firms with higher capital expenditure are affected more by the differential treatment of depreciation in reporting income for tax and financial purposes. Moreover, Intangible assets and equity income are used to control for differential book and tax treatment and for differences in 17

18 reporting earnings in equity. We also control for SIZE and Market-to-book ratio as growing firms may invest more in tax favored assets which generates timing difference in recognizing expenses. Lastly, to control for contemporaneous local macro-economic conditions (see Agrawal and Matsa (2013)), we include state GDP growth rates (based on data from the US Bureau of Economic Analysis) in our regressions. We also include state and industry dummies to control for industry characteristics and overall state characteristics. Table 3 presents the Pearson correlation analysis among the variables used in our analysis. First two measures of tax avoidance proxies (ETR and CETR) are negatively related with second sets of tax avoidance proxies (DATAX and SHELTER). The correlation coefficient between ETR and CETR is positive and significant with a coefficient value of and a p- value of and between DTAX and SHELTER is with a p-value of The correlation coefficients between UnempRisk and ETR, CETR, DTAX, and SHELTER are 0.072, 0.067, , and , respectively, all of which are statistically significant. These positive coefficients for first two and negative coefficients for second two tax avoidance proxies provide initial support to our main hypothesis that higher unemployment risk leads firms to avoid less tax. Most of the correlation coefficients among control variables are small, overriding multicollinearity concerns. 4. Empirical results 4.1. Univariate analysis Our primary objective is to investigate the relation between tax avoidance and labor unemployment risk. The results of our univariate comparison are presented in Table 4. Based on 18

19 the median value of log total UI benefit (UnempRisk), we divide the sample of firm-years into two groups. For the low unemployment risk firm-years, the mean (median) for ETR 0.23 (0.28), CETR 0.17 (0.07), DTAX 0.22 (0.09), and SHELTER is 2.42 (-0.05). For the high benefit group the mean (median) of ETR 0.26 (0.34), CETR 0.20 (0.13), DTAX 0.10 (0.03), and SHELTER is 0.16 (-0.88). The differences in means and medians between the two subgroups are statistically highly significant and this is the case for all the tax avoidance proxies. These findings show that firms with high unemployment risk (i.e., low UI benefits) avoid less tax than those with low unemployment risk (high UI benefits). Our univariate comparison in Table 4 also suggests that firms located in states with higher labor unemployment risk are generally smaller in size, more profitable, have more plants and equipment, have a lower market-to-book ratio, and have higher leverage ratios Multivariate analysis Table 5 presents the regression analysis that tests our hypothesis that firms located in states with higher levels of labor unemployment risk will avoid less tax. 12 We conduct the analyses using four proxies for tax avoidance (e.g., ETR, CETR, DTAX, and SHELTER) and using log of maximum total benefit calculated as the UnempRisk variable. 13 Our findings suggest that higher level of unemployment risk (i.e., a lower value of log of maximum total benefits) is negatively associated with tax avoidance. These results are consistent across all the measures of tax avoidance and are significant at the 1 percent level in all of our regressions. Note that higher 12 Note that in all our regressions in the paper, we adjust standard errors for two-way clustering at the firm level and state level. 13 In later analysis, we corroborate our findings by using Log max weekly benefit as an alternative measure of state UI benefits or unemployment risk. 19

20 values for ETR and CETR (DTAX and SHELTER) represent lower (higher) tax avoidance. So, we do expect to find a positive (negative) association between UnempRisk and ETR or CETR (DTAX or SHELTER). Specifically, we find that the coefficients on UnempRisk variable are positive and significant when ETR and CETR are used (coefficients are and 0.105) and are negative and significant when DTAX and SHELTER are used (coefficients are and ) as tax avoidance proxies. We also run firm fixed effect model in Column (5) to (8) in order to control for unobservable time-invariant firm characteristics. All the coefficients are significant and consistent with those of previous models (column 1 to 4). The estimates in Table 5 (column 1 and 4) show that increases in unemployment risk (UnempRisk) are associated with increase in ETR and CETR and decrease in DTAX and SHELTER. This relation is both statistically significant and economically meaningful. A 100 log point decrease in the log of maximum total benefit (UnempRisk) is associated with 8.2 percentage points (for ETR) and 10.5 percentage points (for CETR) higher than average market ETR and CETR, respectively and with 30.2 points (for DTAX) and prediction score points (for SHELTER) lower than average market ETR and CETR, respectively. We also analyze the economic significance of our findings by multiplying the coefficient estimates for the UnempRisk variable by the change in the level of log maximum total benefit when moving from the 75th percentile to the 25th percentile (Table 2 shows a decrease of (= ). The economic effect of UnempRisk on ETR and CETR can therefore be represented as an increase in tax rates of 4.15 percent (ETR) and 5.31 percent (CETR), respectively. Using the mean level of pre-tax income of $113 million for our sample firms (this number is not reported in any of our tables, but available upon request), this effect translates into an increase between $4.69 million and $6.00 million of tax expenditures, respectively for an average firm-year. 20

21 Overall, the findings of our primary empirical analysis are consistent with the hypothesis that firms located in states with higher unemployment risk (i.e., low UI benefits) are less tax avoiding. Our finding is also consistent with recent literature (e.g., Agrawal and Matsa, 2013) that shows that labor market frictions may have a significant impact on the corporate environment. 5. Additional analysis 5.1. Effect of labor intensity on unemployment risk and tax avoidance Especially in industries where labor intensity is relatively high, one would expect an even stronger relation between unemployment risk and tax avoidance. We test this notion by including a labor intensity variable and an interaction variable of labor intensity and our variable that proxies for unemployment risk in our regressions. As in Pontuch (2011), we measure labor intensity as the inverse of operating assets (data# 6 data# 1) divided by the number of employees (data# 29). 14 We then create a dummy variable (labeled LaborIntensity) that is equal to 1 if the labor intensity is above the median based on two digit SIC code and zero otherwise. Empirically, we estimate the following model: 14 We take inverse of labor intensity ratio for an easy interpretation of regression coefficients. A higher ratio indicates higher labor intensity. Although not reported in a table, the mean (median) labor intensity is 0.012(0.007) and the standard deviation is

22 TaxAvoidance it = α 0 + β 1 UnempRisk st 1 LaborIntensity it + β 2 UnempRisk st 1 + β 3 LaborIntensity it + β 4 ROA it + β 5 LEV it + β 6 NOL it + β 7 NOL it + β 8 FI it + β 9 PPE it + β 10 INTANG it + β 11 EQINC it + β 12 SIZE it 1 + β 13 MB it 1 + β 14 GDP_gr it + StateDummies + IndustryDummies + ε it (4) We expect to find the coefficient on the interaction variable to be positive for ETR and CETR and negative for DTAX and SHELTER. Table 6 presents the results of our analysis. We do find the interaction term to have a significantly positive coefficient when ETR (0.016, t = 1.85) and CETR (0.020, t = 2.22) and a significantly negative coefficient when DTAX (-0.035, t = ) and SHELTER (-2.042, t = -4.05) are used as proxies for tax avoidance. These results suggest that labor intensity may have an effect on the relation between unemployment risk and tax avoidance Effect of financial constraint on the relation between unemployment risk and tax avoidance Our main argument is based on the idea that workers potential job loss or perceived unemployment risk lead them to require higher level of wage differentials. This fear of job loss goes up when a firm is unable to raise external financing needed to mitigate economic crisis. Ofek (1993) concludes that the natural reaction by firm of financial constraint related performance decline is laying off workers. Therefore, increased unemployment risk coming from financial constraint adds up to the existing unemployment risk caused by lower UI benefit. In other words, workers in financially constraint firms will demand even higher wage differential in the event of decreased UI benefit, putting pressure on the firm to maintain conservative corporate 22

23 activities such as tax avoidance. Hence, we expect the relation between unemployment risk and tax avoidance to be more pronounced among financially constraint firms. To explore this possibility, we categorize firm on the basis of a commonly used measure of financial constraintoperating cash flow. Kaplan and Zingales (1997) argue firms with low operating cash flow find it difficult to raise external fund. We divide the sample based on sample median. Firms above the median cash flows face fewer financial constraints as opposed to below the median. Empirically, we test the following equation: TaxAvoidance it = α 0 + β 1 UnempRisk st 1 FinConstraint it + β 2 UnempRisk st 1 + β 3 FinConstraint it + β 4 ROA it + β 5 LEV it + β 6 NOL it + β 7 NOL it + β 8 FI it + β 9 PPE it + β 10 INTANG it + β 11 EQINC it + β 12 SIZE it 1 + β 13 MB it 1 + β 14 GDP_gr it + StateDummies + IndustryDummies + ε it (5) We report the results of this analysis in Table 7 and expect the interaction between UnempRisk and FinConstraint to be positive for ETR and CETR and negative for DTAX and SHELTER. Consistent with our prediction, we do find the interaction term to have a significantly positive coefficient when ETR (0.024, t = 2.54) and CETR (0.013, t = 1.24) and a significantly negative coefficient when DTAX (-0.067, t = -5.15) and SHELTER (-4.007, t = -6.56) are used as proxies for tax avoidance 15. These results strengthen the possibility that the relation between unemployment risk and tax avoidance is not limited to firms facing significant financial constraints. 15 We also classify firms based on their dividend policy. Firms that need to retain their capital are more likely to be financially constraints. All non-dividend paying firms are categorized as financially constrained. In an unreported table, we also find significant negative coefficients for aggressive tax avoidance proxies: DTAX (-0.057, t = -2.31) and SHELTER (-3.060, t = -7.19). 23

24 5.3. Effect of bargaining power on unemployment risk and tax aggressiveness Our main hypothesis states that higher unemployment risk leads to lower tax avoidance. However, especially in industries where employees may have relatively more bargaining power there may be an even stronger relation between unemployment risk and tax avoidance. Prior literature provides substantial evidence that labor unions assess managers investment decisions through the eyes of bondholders, suggesting that unions prefer less risk than shareholders or mangers (e. g., Faleye, Mehrotra, and Morck, 2006; Chen, Kacperczyk, and Ortiz-Molina, 2012; Chyz, Leung, Li, and Rui 2013). An alternative view that union can bargain away a part of firm s after tax profit may lead to higher tax avoidance. However, this argument is reasonable when corporate tax saving activities come with no risk and costs. Therefore, any corporate activity (e.g., tax avoidance) that potentially increases overall risk of the firm is less preferred by unions. In the presence of higher unemployment risk, labor bargaining power may further restrict managers tax avoiding activities. Hence, we expect the relation between tax avoidance and unemployment risk to be more pronounced in firms with higher level of employee bargaining power. We test this by including a bargaining power variable and an interaction variable of this bargaining power and the unemployment risk in the regressions. Our bargaining power variable is defined as the state unionization rate and is collected from the Bureau of Economic Analysis. Empirically, we estimate the following model: 24

25 TaxAvoidance it = α 0 + β 1 UnempRisk st 1 BargPower it + β 2 UnempRisk st 1 + β 3 BargPower it + β 4 ROA it + β 5 LEV it + β 6 NOL it + β 7 NOL it + β 8 FI it + β 9 PPE it + β 10 INTANG it + β 11 EQINC it + β 12 SIZE it 1 + β 13 MB it 1 + β 14 GDP_gr it + StateDummies + IndustryDummies + ε it (6) We expect to find the coefficient on the interaction variable of bargaining power to be positive for ETR and CETR and negative for DTAX and SHELTER. Table 8 presents the results of estimating Equation (6). Consistent with our priors, we find the interaction term to have a significant positive coefficient when ETR (0.024, t = 2.54) and CETR (0.013, t = 1.24) and a significantly negative coefficient when DTAX (-0.067, t = -5.15) and SHELTER (-4.007, t = ) are used as proxies for tax avoidance. Overall, we interpret the results of this analysis as providing evidence that labor bargaining power may have an effect on the relation between unemployment risk and tax avoidance. 6. Robustness checks 6.1. Unobservable firm characteristics and outlier observations In order to address unobservable time-invariant firm characteristics, we showed the results of running a firm fixed effect model (Equation (3)) in Table 5 and our main results did not change. The impact of unemployment risk on tax avoidance is also robust to excluding outlier observations. To test this, we collapse the sample into state-year observations and analyze tax 25

26 avoidance by estimating a median regression. 16 In an unreported table (available upon request), the coefficients on UnempRisk for all measures of tax avoidance are statistically significant and have similar signs as in previous analyses. This suggests that our results are not driven by outliers Controlling for global firms and R&D intensive firms We also run several other variations of the base regression model (Equation (3)) to investigate the robustness of our findings. One concern is that our sample may have global firms, for whom the headquarter state does not capture the location of employees very well. To investigate this possibility, we replicate Equation (3) with all control variables, but exclude global firms. We identify global firms as firms with positive foreign sales reported in the Compustat segment filing. 17 The results are reported in Panel A of Table The results are qualitatively the same for all measures of tax avoidance and are similar as those obtained in our earlier analyses. Another potential concern is the effect of R&D on ETR. Specifically, in hi-tech firms, which tend to have relatively high R&D, this may be an issue as high levels of R&D expenses may cause ETR to be lower. Following financial accounting rules, only acquired intangibles are reported on the balance sheet. However, the R&D expenses could proxy for research and experimentation tax credit. In order to address this issue, we re-estimate Equation (3) using two 16 We take median values when collapsing into state-year observation. As a result, the sample size gets reduced. In median regression, standard errors are calculated using the bootstrap method. 17 We use geography as the segment type while identifying global firms. Therefore, if a firm reports foreign sales along with domestic sales, it is considered to be a global firm. 18 Note that for brevity in this Table and Table 10 we do not report coefficients on control variables. 26

27 approaches: First, excluding R&D intensive industries (e.g., pharmaceuticals, aircraft, motor vehicle, and computer electronic equipment- SIC 283, 371, 372, 376, 357, 384, and 873) from our sample; Second, using an R&D intensity control variable, which is calculated as R&D expense divided by total assets. 19 We report the results of the first approach in Panel B of Table 9 and of the second approach in Panel C of Table 9. For all the tax avoidance proxies (ETR, CETR, DTAX, and SHELTER), we find further consistent evidence that the association between unemployment risk and tax avoidance is negative Controlling for loss firms, big firms, and firms in dispersed industries Finally, we investigate whether our results differ when we exclude loss firms and big firms. The reason to exclude loss firms (defined as firms with negative ROA) is that loss firms are more likely to have higher tax loss carry forwards, which together with current year losses reduce the tax expense for firms. Therefore, it is possible that we erroneously identify a firm as avoiding more tax. The effect of size on firm s effective tax rate is inconclusive. Siegfried (1972) argues that larger firms are likely to be more tax aggressive than smaller firms because they possess greater economic and political power and have more resources to devote in tax planning. However, Stickney and McGee (1983) finds no convincing evidence that size plays a dominant role in explaining differences in effective tax rates. In order to address the concern that bigger firms might have greater impact on tax avoidance, we exclude big firms (defined as total assets greater than the sample median). The results on our regressions of excluding loss firms are reported in Panel A of Table 10 and excluding big firms are reported in Panel B of Table We delete observations where R&D expense is missing. 27

28 Similar to virtually all of our previous analyses, it is clear that there remains a statistically significant relation between unemployment risk and tax avoidance (measured by ETR, CETR, DTAX, and SHELTER). Workers receive UI benefits from the state in which they work. However, our empirical analysis measures UI benefit in the state where their companies are headquartered. This may contaminate our estimates if workers work in a state different from state that their companies are headquartered in. To address this issue, we estimate Equation (3) after excluding industries in which a large portion of labor force is likely to be dispersed (e.g., retail, wholesale, and transport). The results are presented in Panel C of Table 10. As expected, the coefficient of UempRisk are consistent with those in previous analyses. The results further support our main hypothesis that firms with higher unemployment risk avoid less tax Alternative measures of unemployment risk We also use the logarithm of the maximum weekly benefit amount (Log max weekly benefit) as an alternative proxy for unemployment risk. A higher maximum weekly amount should reduce the risk and costs of joblessness. We estimate Equation (3) and, as expected, find that the coefficient on UnempRisk is positive and significant when ETR and CETR are used and negative and significant when DTAX and SHELTER are used. 20 Overall, the findings are consistent with the hypothesis that firms located in states with higher unemployment risk (i.e., lower UI benefit) avoid less tax. 20 The results are not reported here, but available upon request. 28

29 7. Summary and conclusions We examine, and find evidence consistent with, the claim that labor unemployment risk is negatively related to corporate tax avoidance. Recent literature suggests that labor unemployment risk affects a firm s corporate decisions. And, given that tax avoidance is risky and costly for the firm, we argue that higher UI benefits decrease labor unemployment risk and, hence, provide firms with incentives to avoid more tax. Consistent with this hypothesis, we provide empirical evidence of a negative association between labor unemployment risk and tax avoidance (i.e., a positive association between state UI benefits and tax avoidance). Our results are economically significant. In additional analysis, we also find that this negative association is more pronounced for firms in industries that are more labor intensive, firms with financial constraints, and industries with more labor bargaining power. In conclusion, our paper suggests that labor unemployment risk plays an important role in firm s tax policy. 29

30 Appendix A Variable definitions Variable Definition Formula in Compustat items ETR Total tax expense divided by pretax income (data16)/data170 CETR Cash taxes paid divided by pretax income (data317)/data170 DTAX Residual from following equation estimated by year and 2-digit SIC code: PERMDIFF it=α 0+ α 1INTANG it+ α 2UNCON it+ α 3MI it+ α 4CSTE it+ α 5NOL it+ α 6LAGPERM it+є it Where: PERMDIFF = Total book-tax differences temporary book-tax differences = [{PI- [(TXFED + TXFO)/STR]}- (TXDI/STR)], scaled by total assets at year t-1; STR = Statutory tax rate; UNCON = Income (loss) reported under the equity method divided by total assets at year t-1; MI = Income (loss) attributable to minority interest, scaled by total assets at year t-1; CSTE = Current state tax expense, scaled total assets at year t- 1; NOL = Change in net operating loss carry forwards, scaled by total assets at year t-1; LAGPERM = PERMDIFF in year t-1 SHELTER Shelter prediction score Wilson (2009) SHELTER = BTD LEV SIZE ROA FOR_INCOME R&D where: BTD is book income less taxable income scaled by lagged total assets. Book income is pre-tax income. Taxable income is calculated by grossing up the sum of the current federal tax expense and the current foreign tax expense and subtracting the change in NOL carryforward. If the current federal tax expense is missing, then total current tax expense is calculated by subtracting deferred taxes, state income taxes and other income taxes from total income taxes; UnempRisk LEV is long-term debt divided by total assets; SIZE is the log of total assets; ROA is pre-tax earnings divided by total assets; FOR_INCOME is pretax foreign income divided by lagged total assets; R&D is divided by lagged total assets Natural log of the product of max weekly benefit and max duration that a worker is entitled to receive as state UI benefits. Where: Log max weekly benefit is the natural log the maximum weekly wage benefit allowance given to workers in an average stateyear; Log max duration is the natural log of the maximum number of weeks that a state provides unemployment insurance benefits to claimants PERMDIFF = [data170 ((data63 + data64)/ 0.35)-(data50/0.35)]/lag data6 STR=0.35 UNCON=data55/lag data6 MI=data49/lag data6 CSTE=data173/lag data6 NOL=data52/lag data6 book income=data170 taxable income=((data63 + data64)/0.35)-data52 if data63 is missing, it is equal to data16-data50-data173- data211; data50, data173, data211 set to 0 if missing data52 set to 0 if missing LEV=data9/data6 SIZE=log(data6) ROA=data170/data6 FOR_INCOME=data273/lagged data6; 0 if missing R&D=data46/lagged data6 ROA Operating income scaled by lagged assets (data170-data192)/lagged data6 LEV Long-term debt scaled by lagged assets data9/lagged data6 NOL Loss carry forward scaled by lagged assets data52/lagged data6 30

31 NOL Change in net operating loss carry forwards, scaled by total assets at year t-1 FI Foreign income scaled by lagged assets data273/lagged data6 PPE Net property, plant, and equipment scaled by lagged assets data8/lagged data6 INTANG Intangible assets scaled by lagged assets data33/lagged data6 EQINC Equity income in earnings scaled by lagged assets data55/lagged data6 SIZE Natural log of market value of equity at the beginning of year data199 data25 MB Market value of equity divided by book value of equity at the (data199 data25)/lagged data60 beginning of the year. GDP_gr Change in state GDP in year t and t-1 LaborIntensity Operating assets divided by the number of employees (data6-data1)/data29 FinConstraint Below median operating cash flows are financially constrained firms (data18+data14) 31

32 References Abowd, J. M., and O. C. Ashenfelter Anticipated unemployment, temporary layoffs, and compensating wage differentials. Studies in labor markets. University of Chicago Press: Agrawal, A. K., and D. A. Matsa Labor unemployment risk and corporate financing decisions. Journal of Financial Economics: forthcoming. Armstrong, C. S., J. L. Blouin, and D. F. Larcker The incentives for tax planning. Journal of Accounting and Economics 53(1): Badertscher, B. A., S. P. Katz, and S. O. Rego The separation of ownership and control and corporate tax avoidance. Journal of Accounting and Economics 56(2): Benmelech, E., N. K. Bergman, R. J. Enriquez Negotiating with labor under financial distress. Review of Corporate Finance Studies 1(1): Berk, J. B., R. Stanton, and J. Zechner Human capital, bankruptcy, and capital structure. Journal of Finance 65(3): Brown, J., K. Drake, M. Martin Are Firms Myopic?: The Case of Contracting on Performance and Uncertainty. Working Paper. Arizona State University. Brown, J. and D. A. Matsa Boarding a sinking ship? An investigation of job applications to distressed firms.. Working Paper. Northwestern University. Chen, H. J., M. Kacperczyk, H. Ortiz-Molina Labor unions, operating flexibility, and the cost of equity. Journal of Financial and Quantitative Analysis 46(01): Chen, S., X. Chen, Q. Cheng, and T. Shevlin Are family firms more tax aggressive than non-family firms? Journal of Financial Economics 95(1): Chyz, J. A., W. S. Ching Leung, W. S. Zhen Li, and R. Meng Labor unions ands tax aggressiveness. Journal of Financial Economics 108(3): Crocker, K. J., and J. Slemrod Corporate tax evasion with agency costs. Journal of Public Economics 89(9): Desai, M., and D. Dharmapala Corporate tax avoidance and firm value. Review of Economics and Statistics 91: Desai, M. A., and D. Dharmapala Corporate tax avoidance and high-powered incentives. Journal of Financial Economics 79(1): Desai, M. A., and D. Dharmapala Tax and corporate governance: an economic approach. Tax and corporate governance, Springer: Desai, M. A., A. Dyck, and L. Zingales Theft and taxes. Journal of Financial Economics 84(3): Diamond, P. A Aggregate demand management in search equilibrium. Journal of Political Economy. 90(5): Dyreng, S. D., M. Hanlon, and E. L. Maydew Long-run corporate tax avoidance. Accounting Review 83(1): Faleye, O., V. Mehrotra, and R. Morck When labor has a voice in corporate governance. Journal of Financial and Quantitative Analysis 41(3): 489. Frank, M. M., L. J. Lynch, and S. O. Rego Tax reporting aggressiveness and its relation to aggressive financial reporting. Accounting Review 84(2): Gormley, T., H. Liu, and G. Zhou Limited participation and consumption-saving puzzles: A simple explanation and the role of insurance. Journal of Financial Economics 96(2):

33 Graham, J. R., and A. L. Tucker Tax shelters and corporate debt policy. Journal of Financial Economics 81(3): Guenther, D. A., S. R. Matsunaga, and B. M. Williams Corporate tax aggressiveness and firm risk. Working Paper. University of Oregon. Guenther, D. A., E. L. Maydew, and S. E. Nutter Financial reporting, tax costs, and booktax conformity. Journal of Accounting and Economics 23(3): Gupta, S., and K. Newberry Determinants of the variability in corporate effective tax rates: Evidence from longitudinal data. Journal of Accounting and Public Policy 16(1): Hamermesh, D. S., and J. R. Wolfe Compensating wage differentials and the duration of wage loss. Journal of Labor Economics 8: S175-S197. Hanlon, M., and S. Heitzman A review of tax research. Journal of Accounting and Economics 50(2): Jahoda, M Employment and unemployment: A social-psychological analysis. Cambridge University Press, Cambridge, UK. Kaplan, S.N. and Zingales, L Do investment-cash flow sensitivities provide useful measures of financing constraints? Quarterly Journal of Economics 112, Kim, J., O. Li, and Y. Li Tax aggressiveness and bank loan contracting. Working Paper. City University of Hong Kong. Lanis, R., and G. Richardson The effect of board of director composition on corporate tax aggressiveness. Journal of Accounting and Public Policy 30(1): Lanis, R., and G. Richardson Corporate social responsibility and tax aggressiveness: An empirical analysis. Journal of Accounting and Public Policy 31(1): Lazear, E. P Firm-specific human capital: A skill-weights approach. Journal of Political Economy 117: Leung, S. C., O. Li, and O. M. Rui Labor Union and Accounting Conservatism. Working Paper. Chinese University of Hong Kong. Li, E. H Compensating differentials for cyclical and noncyclical unemployment: the interaction between investors' and employees' risk aversion. Journal of Labor Economics 4: Lisowsky, P., D. Mescall, G. Novack, and J. Pittman The importance of tax aggressiveness to corporate borrowing costs. working paper. University of Illinois, Urbana-Champaign, IL. Manzon Jr, G. B., and G. A. Plesko The Relation Between Financial Reporting Measures of Income. Tax Law Review 55: Matsa, D. A Capital structure as a strategic variable: Evidence from collective bargaining. Journal of Finance 65(3): Meyer, B. D Unemployment insurance and unemployment spells. Econometrica 58: Meyer, B. D., and W. K. Mok Quasi-experimental evidence on the effects of unemployment insurance from New York State. Working Paper. University of Chicago. Mills, L. F Book-tax differences and Internal Revenue Service adjustments. Journal of Accounting Research 36(2): Mills, L. F., and K. J. Newberry The influence of tax and nontax costs on book-tax reporting differences: Public and private firms. Journal of the American Taxation Association 23(1):

34 Mortensen, D. T Job search and labor market analysis. Handbook of labor economics 2: Mortensen, D. T., and C. A. Pissarides Job creation and job destruction in the theory of unemployment. Review of Economic Studies 61(3): Ofek, E Capital structure and firm response to poor performance: an empirical analysis. Journal of Financial Economics 34, Pontuch, P Labor Intensity and Expected Stock Returns. Working Paper. Université Paris- Dauphine - DRM-CEREG. Rego, S., and R. Wilson Executive compensation, tax reporting aggressiveness, and future firm performance. Working Paper. University of Iowa. Rego, S. O Tax Avoidance Activities of US Multinational Corporations. Contemporary Accounting Research 20(4): Rego, S. O., and R. Wilson Equity risk incentives and corporate tax aggressiveness. Journal of Accounting Research 50(3): Richardson, G., G. Taylor, and R. Lanis The impact of board of director oversight characteristics on corporate tax aggressiveness: An empirical analysis. Journal of Accounting and Public Policy 32(3): Scholes, M. S., M. A. Wolfson, M. Erickson, E. Maydew, and T. Shevlin Taxes and business strategy. Prentice Hall, Upper Saddle River, NJ. Shevlin, T Corporate tax shelters and book-tax differences. Tax Law Review 55: 427. Siegfried, J. J The relationship between economic structure and the effect of political influence: Empirical evidence from the federal corporation income tax program (Vol. 2). University of Wisconsin--Madison. Stickney, C. P., and V. E. McGee Effective corporate tax rates the effect of size, capital intensity, leverage, and other factors. Journal of Accounting and Public Policy 1(2), Titman, S The effect of capital structure on a firm's liquidation decision. Journal of Financial Economics 13(1): Topel, R., and F. Welch Unemployment insurance: Survey and extensions. Economica 47(187): Topel, R. H On layoffs and unemployment insurance. American Economic Review 73(4): Topel, R. H Equilibrium earnings, turnover, and unemployment: New evidence. Journal of Labor Economics 2: Weisbach, D. A Ten truths about tax shelters. Tax Law Review 55: Wilson, R. J An examination of corporate tax shelter participants. Accounting Review 84(3):

35 Panel A: Panel D: Panel B: Panel E: Panel C: Panel F: Figure 1. The figures represent the quartiles of increase in state maximum unemployment insurance (UI) benefit by quinquennium over 1987 to Larger increases in benefits are shown in darker shades. Maximum total benefit is the product of the statutory maximum weekly UI benefit and the maximum duration. [These figures are replications of Agrawal and Matsa (2013) using our data in order to validate our datasets and graphically present the evolution of UI benefits over time across states.] 35